teal-sea / zeta-labstate of record · compiled 14 Aug 2026 · revision 9ebdea0 · source

Library · docs/11-f1-and-the-missing-geometry.md

F1 and the Missing Geometry

3,305 words · 285 lines · source

A companion to docs/09-new-ontologies.md. That document surveyed the "new objects" landscape and docs/10-trace-formulas-and-connes.md slowed down on the trace-formula corner; this one slows down on the strangest corner: the geometry that is supposed to sit underneath the integers, and the "field with one element" nobody can define but everybody keeps finding fingerprints of.

The short version

The one proven Riemann Hypothesis — the finite-field version, Deligne 1974 — worked because three things existed: an actual space, an actual operator on it (Frobenius), and an actual positivity theorem pinning the operator's eigenvalues to the critical value. For the integers, all three are missing. The organising dream: Spec Z — the geometric object whose points are the primes — should be a curve over a mysterious base F1, "the field with one element," so Weil's template can run. A literal field with one element cannot exist (a field's axioms require 0 != 1), so every attempt is really an attempt to rewrite the rulebook for "field" and "geometry" until the object does exist — the same move that produced schemes and étale cohomology last time. Below: the recap, the dream, the attempts (Tits, Soulé, Connes–Consani, Borger, Deninger), and the honest scorecard. Every attempt is a PROGRAM, not a result. None is close, by its own authors' account.

1. The template that worked once

Recall the finite-field story from docs/09-new-ontologies.md, in three beats.

First, there was a space. Counting solutions of y^2 = x^3 - x modulo a prime p sounds like arithmetic, not geometry — the "solution set" is a scatter of residues, not a surface you could sand. Grothendieck's rebuild of the 1960s changed what space means: a space is anything that obeys the rulebook. Under the new rulebook the scatter of residues is a curve, and — the part that matters here — Spec Z, the set of primes wearing a geometric structure, is also a legitimate space, a kind of one-dimensional curve whose points are the prime numbers.

Second, there was an operator. Over F_p there is a map you get for free: x -> x^p (Frobenius), whose fixed points are exactly the solutions defined over F_p. Étale cohomology attaches to the curve a finite list of honest finite-dimensional vector spaces, and Frobenius acts on them as an honest matrix. The Lefschetz fixed-point formula converts counting solutions into taking traces of a matrix — the same move as modal analysis, where a messy time signal becomes a clean sum over eigenvalues. Counting became linear algebra.

Third, there was a positivity. Symmetry alone (the functional equation, supplied by duality) never pins eigenvalues down; docs/08-why-it-is-hard.md proves that with counterexamples. The closing input in Weil's proof for curves is the Castelnuovo–Severi inequality — a Hodge-index-type statement, a cousin of "this quadratic form is definite" — and the crucial staging detail is where it lives: to prove RH for the curve C, you work on the square C x C and apply positivity to the graph of Frobenius. (Deligne's general 1974 proof took a different analytic route, but still symmetry-plus-positivity at bottom; docs/09, §1.3.) The output: the Frobenius eigenvalues have absolute value exactly sqrt(p) — all zeros on the critical line, as a THEOREM.

Why "eigenvalue modulus" and "critical line" are the same sentence. This is the hinge of the whole analogy and it is worth doing once with numbers rather than in words. The zeta function of a curve over F_p is a function of T, and one substitutes T = p^{-s} to compare it with a Dirichlet series. Its zeros sit at T = 1/alpha for alpha a Frobenius eigenvalue, so

    p^{-s} = 1/alpha   =>   p^s = alpha   =>   s = log(alpha) / log(p)
    Re(s) = log|alpha| / log(p)

and therefore Re(s) = 1/2 is |alpha| = p^(1/2), not an analogue of it. Take y^2 = x^3 + 3x + 5 over F_101 (zeta.finitefield, dps = 20):

    a_p = -13
    alpha = -6.5 +/- 7.664854858 i          |alpha| = 10.0498756211 = sqrt(101)
    alpha * alphabar = 101   exactly        alpha + alphabar = -13
    s = log(alpha)/log(p) = 0.5 +/- 0.492759299573 i     ->  Re(s) = 0.5

and the Lefschetz formula N_n = p^n + 1 - (alpha^n + alphabar^n) against brute-force counting over F_{p^n}: 115, 10235, 1028560 for n = 1, 2, 3, both ways. That is the whole payoff of item two — counting became a trace — and item three is what forces the modulus.

The reason this matters for the rest of the document: a constraint on where a zero sits is fragile and analytic, while a constraint on how big an eigenvalue is is structural, and structural constraints have proofs. Every program below is trying to buy that change of coordinates for zeta. scripts/11_finite_field_rh.py runs the case where it works.

Plain-words recap. The only RH ever proved needed three ingredients: a space, a matrix acting on something built from it, and a positivity theorem — applied on the space's square — forcing the eigenvalues onto the critical value. Space, operator, positivity. Keep the checklist; the rest of this document is people trying to build item one. And note why eigenvalues are the prize: s = log(alpha)/log(p) turns "the zero is on the critical line" into "the eigenvalue has modulus sqrt(p)" — the same statement in better coordinates.

2. The dream: Spec Z as a curve over F1

Here is the gap, stated as an engineering problem. Weil's proof runs on C x C, the square of the curve — and "square" means square over the base field: a curve over F_p is a one-parameter object with F_p as its fixed ground, so the product is a genuine two-parameter object, a surface. Try the same with the integers. What is the ground under Z? Nothing: Z is the bottom of the tower, and the product collapses — Z (x)_Z Z = Z, a theorem-level fact (docs/09, §4.1). The square of Spec Z over its only available base is just Spec Z again: a one-dimensional object whose square is one-dimensional. The staging area for the positivity argument does not exist.

The dream postulates a basement. Suppose there were a field F1 below Z — a "field with one element" — such that Spec Z is a curve over F1. Then Spec Z x_{F1} Spec Z would be an honest surface, Weil's positivity argument would have its home, and the finite-field template could run on the actual Riemann zeta function. The shape of the dream is completely precise. Only the object is missing.

And here is the charming catastrophe: a literal field with one element cannot exist. A field is a number system where you can add, subtract, multiply, and divide by anything nonzero; its axioms require a zero 0 and a unit 1 with 0 != 1 — written into the definition, because if 0 = 1 everything collapses to 0. The smallest field has two elements; "F1" is a contradiction in terms. The response is not to give up but to make the objects-and-rulebooks move of docs/09, §2: if the object you need is forbidden by the current definition, change the definition — carefully, so everything old survives — until the object exists. You have seen this move before. No real number squares to -1; mathematics rebuilt "number" until i existed. No function is zero everywhere, infinite at a point, with integral 1; analysis rebuilt "function" (distributions) until the delta function existed. F1 asks for the same manoeuvre one level deeper: rebuild "field" — or rebuild "geometry" so it no longer needs a field at the bottom.

Why believe the basement is there at all? Because its fingerprints keep showing up.

Plain-words recap. The proof template needs the square of Spec Z, and over the integers the square collapses — there is no ground field under Z to take it over. The dream: invent the missing ground, F1. A one-element field is flatly impossible under the current axioms (0 != 1), so every F1 program is really a proposal to redefine "field" or "geometry" — the same legal move that gave us i and the delta function.

3. The fingerprints, and the attempts

All three items below (and Deninger's program in Section 4) are labelled PROGRAM, per the house convention: genuine theorems inside their own formalisms, no proof of RH, none claimed.

Tits' observation (1950s) — the original fingerprint. PROGRAM. Many counting formulas over the field F_q are polynomials in q, and something eerie happens at q = 1. The projective plane over F_q has q^2 + q + 1 points: 7 at q = 2, 13 at q = 3 — and 3 at the forbidden value q = 1: exactly a triangle, the degenerate "projective plane" with three points and three lines. Deeper: the number of complete flags in F_q^n (a nested chain line-inside-plane-inside-space) is the q-factorial [n]_q! = (1)(1+q)(1+q+q^2)... — at n = 3, q = 2 that is 1 * 3 * 7 = 21, a count you can check by hand: F_2^3 has 7 lines, each lying in exactly 3 of its 7 planes, and 7 x 3 = 21 — and at q = 1 it becomes n!, the number of permutations of n things. Tits noticed this pattern across the algebraic groups: at q = 1 the geometry degenerates into pure combinatorics (Weyl groups — finite symmetry skeletons). As if every F_q-geometry were a q-parameter family whose q -> 1 limit is the geometry over a one-element field. The limit of the formulas exists; the object they count at q = 1 is what nobody can exhibit.

Soulé (2004) and Connes–Consani — rulebook rewrites. PROGRAM. Direct attempts to write the new definition. Soulé proposed a candidate definition of "variety over F1" in the category-theoretic style: since an object is completely determined by how it maps to and from other objects (docs/09, §2), you can define an F1-object by prescribing its interactions — roughly, what its sets of points over every actual field must be — without ever saying what it is made of. Connes and Consani have developed several successive formalisms in the same spirit (some deliberately weakening "addition," since F1 should have a 1, a 0, and essentially nothing else). The honest state of play, unchanged from docs/09: the formalisms genuinely differ, each proves real theorems internally, and none has produced the thing the dream needs — the surface Spec Z x_{F1} Spec Z with a workable intersection theory, the staging area for positivity.

Borger's lambda-rings — F1-structure as symmetry. PROGRAM. Borger's approach (commonly cited from around 2009; I have not verified the date) is the most engineer-friendly: it identifies "being defined over F1" not with a new kind of point-set but with extra structure carried by the ring — a lambda-ring structure, which for our purposes means a commuting family of maps lifting the Frobenius x -> x^p for every prime p at once. Read it as a design decision: last time the operator came for free from the space; this time, since we cannot find the space, declare the operator to be the structure — "Frobenius over the integers" becomes the definition rather than the mystery. The integers carry this structure canonically, which is the program's founding observation; whether it yields the cohomology and the positivity is open.

Plain-words recap. The evidence for the basement: formulas keep having sensible values at the illegal setting q = 1 — geometry degenerates to counting permutations. The attempts: define F1-objects by their interfaces instead of their insides (Soulé, Connes–Consani), or skip the space and axiomatise the operator — F1-structure as "all Frobenius symmetries at once" (Borger). Real mathematics in each; the load-bearing surface still missing in all.

4. Deninger's program: the primes as a dynamical system

Deninger's program (from the early 1990s; ICM address commonly cited as 1998) approaches the missing object from the physics side, and it is the one this repository has been secretly preparing you for.

Start from what you already know. In docs/04-explicit-formula.md, every non-trivial zero rho = beta + i gamma is literally a pole of -zeta'/zeta, and in the variable u = log x it contributes a mode e^{beta u} cos(gamma u - phase) to the prime-counting error — exactly what a pole at s = beta + i gamma contributes to the impulse response of a linear system: gamma the ringing frequency, beta the growth exponent. After the natural sqrt(x) = e^{u/2} normalisation, RH is a pole-placement statement: every pole of the system sits exactly on the marginal-stability line — every mode a pure, undamped, ungrowing oscillation, none hot, none dead. And on the other side of the same formula, the primes enter as the sequence log p, 2 log p, 3 log p, ... — for p = 2: 0.6931, 1.3863, 2.0794, ... — precisely the signature of a closed orbit of period log p and its repeated traversals, the way periodic orbits enter trace formulas in quantum chaos (docs/06-hilbert-polya-and-gue.md, the Gutzwiller connection).

Deninger's proposal takes that reading literally. PROGRAM: there should exist an actual dynamical system — a flow on some infinite-dimensional foliated space (layered into lower-dimensional leaves, like plywood, the flow moving along the leaves), none of it yet constructed — such that: the closed orbits of the flow are the primes, with period log p; the flow acts on an infinite-dimensional cohomology H^1 (the analogue of Section 1's étale spaces) with generator Theta; zeta is — up to elementary factors from the companion spaces H^0 and H^2, which carry its pole — the regularized determinant det(s - Theta) over that H^1 (a determinant of an infinite matrix, made finite by regularisation), so the zeros are exactly the eigenvalues of Theta — the modal frequencies of the flow; and the explicit formula of docs/04 becomes a Lefschetz-type trace formula: spectrum on one side, periodic orbits on the other, an identity by geometry instead of contour integration. Deninger has built real fragments — e.g. determinant formulas reproducing the Gamma factor of the functional equation (docs/09, §4.3; I state the fragments at that document's level of confidence and no further). The space carrying the flow has not been found.

Notice how this rhymes with docs/05-de-bruijn-newman.md. The de Bruijn–Newman heat flow is a genuine dynamics with the exact critical behaviour RH demands (Lambda = 0 iff RH) — but it acts on the function Xi, a dynamics with no space underneath it. Deninger predicts the space: a flow whose orbits are the primes, whose spectrum is the zeros, and for which a positivity — the third checklist item — could finally be a statement about an actual system, the way "this structure's modes are all real" is a statement about a symmetric stiffness matrix. The F1 programs of Section 3 and Deninger's triangulate the same missing object from different sides: primes as points of a curve, primes as closed orbits of a flow.

Plain-words recap. You already read the explicit formula like an engineer: zeros are poles, RH says all poles sit on one vertical line, and the primes tick in at log p, 2 log p, 3 log p — the calling card of a periodic orbit. Deninger conjectures the machine behind the transfer function: a flow whose closed orbits are the primes and whose modal spectrum is the zeros. Fragments exist; the machine does not, yet.

5. What any of these must deliver, and the scorecard

The four falsification gates of docs/09, §5, applied to this document's programs:

The scorecard, soberly. The precedent cost roughly twenty years, a generation of exceptional people, and thousands of pages of new foundations — with a working blueprint the whole time, since Weil had already proven the curves case. Round two has the same blueprint and, counting from Tits' remark, more than fifty years of searching without finding the objects. In that time: real theorems inside each formalism, genuine restatements of RH in new languages (Connes' trace-formula program restates it as positivity on a space that genuinely exists — docs/09 §4.2, and in full docs/10-trace-formulas-and-connes.md §5), fingerprints of F1 everywhere — and no space, no operator with a home, no positivity. That is not an argument the search will fail. It is a measurement of the size of the ask: last time the answer was a rebuild of what geometry is, and there is no reason to expect round two to be cheaper.

Plain-words recap. Four tests: reproduce the known results; give positivity somewhere to live; make the known counterexample fail to fit; make the primes structurally the points or orbits. Fifty-plus years in, every serious program passes the fourth by design and is stuck on the first two. The precedent says the missing step is a foundations rebuild, not a clever estimate — and rebuilds are bought in decades.

Where to go next